Table of Contents
Hey everyone,
So this newsletter is late because I spent a portion of Friday, and then what bled into my weekend, in a weird AI-fueled time warp, recreating 1985 Halloween vibes. I started with "let me just try this new Veo feature real quick" and ended up elbow-deep in disposable camera aesthetics at 2 AM on Sunday morning.
This is apparently what my life has become. 🙃
But to be fair, it wasn't ALL screen time. We also took baby James to the park, where there was live music about bunnies, and he got to hang out with his little classmates and their parents. Although they couldn’t speak, they were definitely vibing.
Then Saturday I crushed some NA beers with Joe Chura from Go Brewing at Soft Bar in Greenpoint, because apparently my idea of a wild time now involves discussing hop profiles while staying completely sober.
Sunday featured a nice picnic in Prospect Park where Linus nearly caught a squirrel, then mistook another small dog for a squirrel and went full predator mode before doing a u-turn when he realized his mistake. The other dog owner definitely aged five years in those three seconds (RIP), but this was during off-leash hours so technically everyone knew what they signed up for.
Anyway, between normal human activities, I somehow created an entire Halloween campaign.
Veo 3 Gets 9:16 and I Lost My Mind
Google quietly dropped 9:16 aspect ratio support for Veo 3 last week.
This is genuinely huge for anyone making social content. No more awkward letterboxing. No more "this would be perfect if it wasn't the wrong shape." We can finally make stuff that actually fits where people watch it.
But here's the thing nobody talks about: having the right aspect ratio doesn't magically make your content good. I still had to figure out how to not make garbage.
It’s also worth noting that Veo 3 Fast is unlimited with an Ultra account (I believe it’s around $130/month).
The Systematic Approach (That Took Me Way Too Long to Figure Out)
I'm about to walk you through exactly how I made this, mistakes and all, because apparently I document my failures now.
Phase 1: Establish Your Creative North Star
The biggest mistake I see people make is jumping straight into generation without knowing what they're trying to achieve. I started by finding one reference photo on Pinterest that perfectly captured the vibe I wanted - that grainy, disposable camera, 1980s Halloween energy.
Then I used imageprompt.org/image-to-prompt to reverse-engineer what made that photo work. This tool basically tells you all the technical details you're too artistically challenged to notice yourself.
Phase 2: Lock In Your Visual DNA
Before generating anything, I established non-negotiable aesthetic rules and fed them to Claude
I'm building a 1980s Halloween campaign with these non-negotiable visual rules:
- Heavy film grain (always)
- Oversaturated warm tones
- Disposable camera imperfections
- Yellow film date stamps
- Suburban Halloween energy, never urban or modernThis became my creative guardrails. Every subsequent prompt had to honor these rules.
Phase 3: Generate Foundation Scenes
With my aesthetic locked down, I created base scenes in Midjourney using this template:
The very important thing here is to ALWAYS have the object you wish to replace in the scene
A candid 1980s Halloween night snapshot of [SPECIFIC SCENE]. Two adults in their late 20s dressed in [COSTUME 1] and [COSTUME 2] on a suburban [LOCATION]. Streetlamps create harsh yellow highlights, jack-o'-lanterns glow on porches, kids in costumes trick-or-treat in background. Disposable-camera look: heavy film grain, slight blur, oversaturated oranges, underexposed shadows, yellow film date stamp in corner. Authentic 1980s Kodak 400 film with imperfect exposure.I tested multiple costume combinations and settings to find what worked best:
Top Gun + Madonna
Ghostbusters + Prince
Indiana Jones + Cyndi Lauper
The key was generating variations systematically, not randomly.
Two adults in their late 20s dressed in 80s costumes one as Madonna with lace gloves, one as a Ghostbuster with proton pack holding a a yellow beer 12 oz beer can chatting on a decorated porch surrounded by glowing jack-o’-lanterns holding . One leans on the railing while the other gestures mid-sentence. Shot feels unposed and imperfect: slightly off-center, film grain, oversaturated colors, underexposed shadows, yellow film date stamp in corner. Disposable-camera look: heavy film grain, oversaturated oranges and purples, uneven flash lighting, underexposed shadows, film date stamp in the corner.Two 30 year olds pose casually at the end of a suburban driveway during 1980s Halloween. One is dressed as Prince in a purple jacket with ruffles, the other as Cyndi Lauper with neon layers, big hair, and mismatched accessories. Each holds a yellow 12oz beer can, while surrounded by neighborhood activity children in costumes, glowing jack-o’-lanterns, fallen leaves. Shot feels unposed and imperfect: slightly off-focus, grainy, oversaturated oranges and purples, with a film date stamp in the corner. Kodak film look with that distinctive 80s snapshot quality - underexposed in places, overexposed highlights. Fallen autumn leaves on sidewalk, jack-o'-lanterns on porches, plastic Halloween decorations. Busy neighborhood Halloween night feeling. Candid moment captured mid-conversation, not posed. Film date stamp in corner
Also more specific you get about the film grain and date stamp, the more authentic it looks. I don't know why this works, but it does.
Phase 4: Product Integration Surgery
Here's where most people mess up: they try to force their product into Midjourney.
Don’t.
Do.
That.
I learned to test product fit early. If the swap looked forced, I went back to Phase 3 and created a different base scene.
Nano is magic with product insertion, but you have to be surgical about it:
Replace [OBJECT IN HAND] with [YOUR PRODUCT] while maintaining exact hand position, angle, and size. Keep all costume details, lighting, Halloween decorations, and 1980s disposable-camera aesthetic. Only change the product - everything else stays identical.
This is where you find out if your product actually fits the scene or if you're just forcing it. Be honest with yourself.
Phase 5: Animation with Intention
Instead of animating everything, I strategically chose which scenes needed motion. Hero product moments got full Veo 3 treatment.
And as always, use JSON prompts here (ensure Claude provides them); Frame to Video and Veo 3 work perfectly fine (and are unlimited).
{
"version": "veo-3",
"duration_seconds": 8,
"fps": 24,
"aspect_ratio": "9:16",
"look": {
"era": "1980s Halloween night",
"palette": "oversaturated oranges, deep blacks, tungsten porch glow with cool shadows",
"film_grain": "heavy, Kodak 400 disposable camera feel",
"lighting": "harsh porch light mixed with jack-o'-lantern flicker",
"aesthetic": "authentic 80s snapshot: uneven exposure, soft edges, yellow date stamp"
},
"shots": [
{
"duration": 8.0,
"prompt": "[DESCRIBE YOUR SCENE] with disposable-camera authenticity: film grain, oversaturated colors, imperfect flash reflections, date stamp visible in corner",
"camera": { "type": "slight handheld wobble", "speed": "slow push-in" },
"ensure": ["product visible and sharp", "no modern elements", "retro vibe maintained"]
}
]
}
The key is in the "ensure" section, specifying what not to mess up, because it will try to be helpful and ruin everything.
Phase 6: Strategic B-Roll
You need filler content to make this feel professional. Generate standalone atmospheric scenes with Midjourney image, and then animate within Midjourney itself:
Candid 1980s Halloween night snapshot of [NEIGHBORHOOD ACTIVITY]. Adults in their 20s-30s in Halloween costumes [DOING WHATEVER] on suburban street/porch. Porch lights glow unevenly, jack-o'-lanterns flicker, autumn leaves everywhere because it's apparently always autumn in advertising. Disposable-camera aesthetic: heavy grain, oversaturated oranges, soft focus, harsh flash, yellow date stamp. Make it look like someone's dad took this photo in 1987 and just found it in a shoebox.
Phase 7: CapCut Cleanup (AKA Making It Less Obvious I'm Not a Real Filmmaker)
Just basic editing: cuts, transitions, maybe some color grading to hide the fact that I have no idea what I'm doing. The AI did the hard work; I just tried not to ruin it.
What Actually Worked (The Uncomfortable Truth)
The breakthrough wasn't any single tool - it was treating this like a real production with systematic creative development.
What worked:
Establishing aesthetic rules upfront and never breaking them
Testing variations systematically, not randomly
Using Claude to maintain creative consistency across the entire project
Being honest about when product integration looked forced
What didn't work:
Rushing the foundation phase (my first attempts looked like AI garbage)
Trying to animate everything (some scenes work better as stills)
Ignoring when the product didn't fit naturally
The reality check: This process takes time. You're not going to create professional-looking content in 20 minutes. But you can produce something that looks like it cost way more than it actually did.
The Reality Check
This isn't replacing real video production. It's not going to win any awards. But it's a way to create multiple variations of branded content without hiring actors, renting costumes, or explaining to your neighbor why you need to borrow their porch for a "quick photo shoot."
The real value is in rapid iteration. Want to test different costumes? Different products? Different scenes? You can generate variations in hours instead of months.
But here's what nobody tells you: you still need good creative instincts. The AI will happily generate technically perfect garbage if you let it. The tools are getting better, but taste is still your job.
TLDR; The Workflow That Emerged
Find visual reference → reverse-engineer with image-to-prompt
Establish aesthetic rules → document them in Claude
Generate base scenes systematically → test multiple variations
Product integration → surgical swaps, test for naturalness
Strategic animation → hero moments get full treatment
B-roll generation → atmospheric filler for completeness
Light editing → basic cuts and color grading in CapCut
Each phase builds on the previous one. Skip steps and the whole thing falls apart.
The Takeaway
This isn't replacing real video production. It's not winning any awards. But it's a way to create multiple branded content variations without hiring actors, renting costumes, or explaining to your neighbor why you need their porch.
The real value is rapid iteration with maintained quality standards. Want to test different products? Different scenes? Different costumes? You can generate variations in hours instead of months.
But here's what nobody tells you: you still need creative judgment. The AI will happily generate technically perfect garbage if you let it. The tools are getting better, but taste is still your job.
That's it for now!
Send me your experiments if you try this workflow. I promise to give you honest feedback about whether it's good or if you need to try again.
Also, if anyone knows why I find this stuff so addictive, please let me know. I have other things I should be doing.
Best,
Will
P.S. - Yes, I realize I just wrote 1,500 words about making fake 1980s Halloween videos. This is my life now. We're all just going to have to accept it.